Carnegie Mellon Robotics Institute
master's thesis, tech. report CMU-RI-TR-12-02, Robotics Institute, Carnegie Mellon University, December, 2011
|Autonomous air vehicles frequently rely on GPS as a primary source of state feedback. However, relying on GPS disallows operation in enclosed spaces, under heavy vegetation or near large obstacles since GPS does not provide sufficient accuracy in these environments. Similar work in GPS-denied navigation uses laser-based odometry, structured-light or places visual markers in the environment. These approaches are not appropriate for operating in unstructured environments, therefore it is necessary to develop a system around more robust vision-based techniques.
This work presents a method for controlling an autonomous, multi-rotor helicopter based on visual odometry which consists of two major sections. First, a method for developing an accurate dynamic model of a multi-rotor helicopter based on a combination of first priciples and emperical data is presented. This modeling technique is used to develop a control system which enables trajectory tracking. Second, a vision-based state-estimation technique is presented along with a hardware implementation that enables execution of the algorithms in real-time on board a small vehicle with strict payload constraints. The methods described are implemented in flight-ready hardware with a minimal weight, power and computational footprint. The system is then evaluated on board a small, eight-rotor helicopter. We successfully demonstrate vision-based trajectory tracking in flight on this vehicle.
Number of pages: 32
|Justin Haines, "Vision-Based Control of a Multi-Rotor Helicopter ," master's thesis, tech. report CMU-RI-TR-12-02, Robotics Institute, Carnegie Mellon University, December, 2011|
author = "Justin Haines",
title = "Vision-Based Control of a Multi-Rotor Helicopter ",
booktitle = "",
school = "Robotics Institute, Carnegie Mellon University",
month = "December",
year = "2011",
address= "Pittsburgh, PA",
|The Robotics Institute is part of the School of Computer Science, Carnegie Mellon University.|
Contact Us | Update Instructions